How to crop image by yolo annotated data?

Question

I am trying to write a python script which will crop image by annotated data. Can anyone help me?

Image:

Annotated data:

0 0.514583 0.716204 0.136458 0.102778

Answer 1

The YOLO coordinates mentioned are in format: x_center, y_center, width_box, height_box, normalised wrt image height and width. You can convert it to normal/usual format for drawing rectangle using:

x = x_center * image_width
y = y_center * image_height
w = width_box * image_width
h = height_box * image_height

If you looking for bounding box format as: xmin, ymin, xmax, ymax, then:

xmin = int(x - width_box/2)
ymin = int(y - height_box/2)
xmax = int(x + width_box/2)
ymax = int(y + height_box/2)

Answer 2

you can do like this >>

import cv2, os

image = cv2.imread('/file/path.ext')
lh,lw,_ = image.shape

# now take your coordinates
x,y,w,h = 0.514583, 0.716204, 0.136458, 0.102778
x,y,w,h = int(x*lw), int(y*lh), int(w*lw), int(h*lh) ## to match the bounding box coordinates with actual width, height

boxedImage = image[y:y+h, x:x+w]
cv2.imshow(boxedImage)

Hope this helps!

Answer 3

You can use this code to load your image and annotation files into the program, crop the images using the annotation coordinates, and save the cropped images as new files.

import cv2
import numpy as np
import os

# Function to display an image using OpenCV
def show_image(image, title='Image'):
    cv2.imshow(title, image)
    cv2.waitKey(0)
    cv2.destroyAllWindows()

# Load the YOLO annotations
annotations_file = r"path/to/annotation.txt"
annotations = np.loadtxt(annotations_file, delimiter=" ")

# Load the image
image_file = r"path/to/image.jpg"
image = cv2.imread(image_file)
print(image.shape)
# Convert the YOLO annotations to pixel coordinates
image_height, image_width, _ = image.shape

x_center = annotations[1]
y_center = annotations[2]
width = annotations[3]
height = annotations[4]
x_min = int((x_center - width / 2) * image_width)
y_min = int((y_center - height / 2) * image_height)
x_max = int((x_center + width / 2) * image_width)
y_max = int((y_center + height / 2) * image_height)

print("x_min:", x_min)
print("y_min:", y_min)
print("x_max:", x_max)
print("y_max:", y_max)
# Crop the image
cropped_image = image[y_min:y_max, x_min:x_max]

# Show the cropped image
show_image(cropped_image)


# Save the cropped image
output_dir = r"path/to/crop_files_dir"
if not os.path.exists(output_dir):
    os.makedirs(output_dir)
output_file = os.path.join(output_dir, "cropped_image.jpg")
cv2.imwrite(output_file, cropped_image)

original image

cropped image